Powerful and developer-friendly framework for building autonomous AI agents.
Combine sub-agents, graph workflows, persistent memory, and MCP-connected tools in one developer-friendly framework built for complex, real-world execution.
Route incoming task
Evaluate role fit, available tools, and memory scope before delegation.
Research Specialist
Pulls product context, prior tasks, and unresolved edge cases.
Policy Specialist
Verifies approval rules, limits, and escalation constraints.
Delegate complex work to specialized agents with graph-aware routing, shared context, and capability-based assignment.
Workflow graph
Dependencies, branching, and task state
queued
08
running
03
complete
21
scheduled
04
Build DAG-based execution flows with dependency management, parallel branches, task status, and scheduled runs.
Memory search
Retrieved context
4,800
tokens injected before response generation
Store scoped memories per session, graph, or task, then retrieve the right context with vector search and automatic injection.
Connected MCP servers
filesystem
local access
github
repo sync
postgres
query layer
active servers
03
scoped attach
task
Tool registry
Connect MCP servers, register plugins, and expose typed tools with schema validation and automatic function calling.
Learn the core APIs, build with the CLI, and move through real-world examples without jumping across disconnected docs.
Framework
Agent.create()
Create agents with memory, knowledge, vision, and orchestration from one typed API surface.
Model layer
OpenAI, Claude, Gemini, Ollama
Built-ins
memory, knowledge, vision, MCP
Core runtime primitives
Agents, tasks, graphs, memory, scheduler, security, vision, knowledge, and MCP live in one consistent framework surface.
CLI
scaffold
agent setup
providers
switch + test
file tools
read, write, edit
Build and iterate in terminal
Use the Astreus CLI to scaffold projects, inspect code, switch providers, and keep development moving from one place.
Examples
Support escalation
Graph-routed support flow with persistent memory, MCP-connected tools, approval-safe escalation logic, reusable context pulled from previous runs, and fallback routing that keeps the workflow stable when confidence drops or tool access changes mid-run.
Example path
Clone the examples repo or install the package and build the same workflow from scratch.
Real working patterns
Start from memory, graphs, sub-agents, plugins, MCP, and scheduling examples that map to real production use cases.
Questions, answered.
The essentials on how Astreus works, what ships in the framework, and where teams use it in real production systems.
Astreus is built for autonomous AI agents, multi-step workflows, sub-agent coordination, memory-backed assistants, MCP-connected tooling, and production task execution from one TypeScript framework.
Yes. Astreus supports specialized sub-agents, delegation, parallel execution, and DAG-style graph workflows with dependencies, conditional logic, scheduling, and task orchestration.
Yes. Persistent memory lets agents store and retrieve context with vector search so they can remember past conversations, reuse relevant details, and maintain continuity over time.
Astreus includes a plugin system and MCP integration for connecting external tools and services. You can register typed tools, validate inputs with schemas, and let agents call them cleanly during execution.
Yes. Agents can work with vision, document processing, and knowledge base features, including RAG-style retrieval, chunking, embeddings, and similarity search for grounded responses.
Astreus exposes a unified interface across providers such as OpenAI, Claude, Gemini, and Ollama, so teams can switch models, compare behavior, and route workloads without rewriting agent logic.
